Advancing Liver Disease Sarcopenia Prediction: A Chest- Radiograph–Based Model for Older Adults

Abstract

Background Sarcopenia has been identified as a significant factor affecting the quality of life and prognosis of ageing population. We have been developing a deep learning model to predict height, weight for appropriate nutritional intervention in bedridden patients based on chest radiographs examinations. In this study, we developed and validated a model for predicting sarcopenia in patients with liver disease based on chest radiographs. The model was developed and validated in patients with liver disease based on chest radiographs. Methods A total of approximately 10842 male and chest Radiograph examinations, including physical examinations conducted in the past 15 years, were subjected to the study. BMI data were used as the teacher data and validation data, and a discriminant model S-CNN was developed using cut off BMI values (18.5, 19, 20). A convolutional neural network ResNet-152 was used to develop the model, and the model was trained on an Nvidia RTX A6000 using Python 3.8 and Pytorch 1.8.1. To validate sarcopenia, we used x-rays of 22 liver disease patients (13 sarcopenia cases) over the age of 65 years, in which grip strength and limb skeletal muscle mass were measured. Limb skeletal muscle mass was measured by bioelectrical impedance (BIA) and skeletal muscle mass index (SMI) was obtained. 2were determined as sarcopenia. Sensitivity, specificity, accuracy, and F1 score were used to evaluate the performance of the deep learning model; Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC) were used to evaluate the prediction accuracy of sarcopenia patients with liver disease. Results The S-CNN model showed a sensitivity, specificity, accuracy, and F 1 scores were 98.0%, 98.0%, 97.0%, and 98%, respectively. Next, the results were validated with chest radiographs including liver disease sarcopenia, AUC = 0.62 (BMI cut off = 18.5), AUC = 0.62 (BMI cut off = 19), and AUC = 0.77 (BMI cut off = 20). Conclusion The highest AUC for prediction of sarcopenia in patients with liver disease by chest radiography was found at BMI (cut off = 20). Development of a more accurate model and its validation on a large scale are expected to realize a simplified pickup of sarcopenia in elderly patients with liver disease.